Abstract

Many children with frontal lobe epilepsy (FLE) have significant cognitive comorbidity, for which the underlying mechanism has not yet been unraveled, but is likely related to disturbed cerebral network integrity. Using resting-state fMRI, we investigated whether cerebral network characteristics are associated with epilepsy and cognitive comorbidity. We included 37 children with FLE and 41 healthy age-matched controls. Cognitive performance was determined by means of a computerized visual searching task. A connectivity matrix for 82 cortical and subcortical brain regions was generated for each subject by calculating the inter-regional correlation of the fMRI time signals. From the connectivity matrix, graph metrics were calculated and the anatomical configuration of aberrant connections and modular organization was investigated. Both patients and controls displayed efficiently organized networks. However, FLE patients displayed a higher modularity, implying that subnetworks are less interconnected. Impaired cognition was associated with higher modularity scores and abnormal modular organization of the brain, which was mainly expressed as a decrease in long-range and an increase in interhemispheric connectivity in patients. We showed that network modularity analysis provides a sensitive marker for cognitive impairment in FLE and suggest that abnormally interconnected functional subnetworks of the brain might underlie the cognitive problems in children with FLE.

Introduction

Frontal lobe epilepsy (FLE) is considered the second most common type of the localization-related (partial) epilepsies of childhood, after temporal lobe epilepsy, and accounts for 20–30% of partial epilepsies (Manford et al. 1992). Pediatric FLE is frequently complicated by cognitive impairment and behavioral disturbances. FLE impacts a broad scale of cognitive domains, broader than the typical frontal functions (Centeno et al. 2010; Braakman et al. 2011). In children with FLE, the learning difficulties may even precede seizure onset, suggesting a nontrivial relationship between seizures and cognitive problems, which has not been unraveled thus far (Prevost et al. 2006; Patrikelis et al. 2009).

The broad scale of affected cognitive domains hints at a network disturbance, rather than disturbance of localized processes. In line with this suggestion, conventional structural MRI studies have not shown an anatomical substrate for the neuronal mechanisms leading to cognitive impairment in FLE (Harvey et al. 1993; Laskowitz et al. 1995; Lorenzo et al. 1995; Provini et al. 1999; Lawson et al. 2002). To resolve this, functional MRI techniques may prove novel and valuable insights. Resting-state fMRI (RS-fMRI) is a functional imaging technique that may be useful in understanding the neuronal mechanisms behind cognitive comorbidity in neurological disorders (Fox and Raichle 2007). RS-fMRI enables the investigation of the intrinsic functional organization of the brain, in contrast to the cerebral effect of tasks executed by subjects. This intrinsic functional organization is called “functional connectivity,” which is defined by the temporal correlation of neuronal activity-induced patterns of anatomically different brain regions (Friston 1994; van den Heuvel and Hulshoff Pol 2010). Previous studies have demonstrated disturbances in functional connectivity networks in the brains of adult epilepsy patients (Waites et al. 2006; Bettus et al. 2009; Liao et al. 2010; Pereira et al. 2010; Vlooswijk et al. 2010, 2011; Pravata et al. 2011). These studies focused on “local” connectivity abnormalities, that is, only a few regions of the brain were analyzed or considered as a reference. Given the heterogeneous expression of cognitive deficits in FLE (Braakman et al. 2011), it is advantageous to investigate the whole-brain network. In mathematical terms, the brain can be modeled as a system consisting of nodes (brain regions) and edges (connections) between them. The strength of a connection is then quantified by the degree of correlation of the dynamic fluctuations between a pair of nodes, that is, brain regions. An elegant way to understand and quantify the organization of this system of nodes and edges is to calculate graph theoretical metrics of the whole-brain network (Rubinov and Sporns 2010). The metrics that provide information on the amount of integration and segregation over the entire brain are path length and clustering. In addition to the whole-brain network organization, graph theoretical metrics can also describe the interconnection of subnetworks in the whole-brain network by modularity analysis (Newman 2006), see Table 1 for a detailed description. The modular structure of the brain network is thought to be important for cognitive abilities as increases in coherent activity between functional systems might facilitate information integration (van den Heuvel et al. 2009) and adaptive behavior (Power et al. 2010).

Table 1

Network parameters used in this study

Measures  Interpretation 
graphic  Characteristic path length is defined as the average geodesic distance, in number of edges, connecting any two nodes in the graph, where di,j is the length of the shortest path between nodes i and j. The characteristic path length is a measure of how well connected a network is. Small characteristic path length indicates an average short distance between any two nodes, that is, they can be reached through a small number of steps. 
graphic  The weighted characteristic path length is defined as the average of the shortest paths connecting any two nodes in the graph, where wi,j is the sum of weights of the shortest weighted path between nodes i and j. The characteristic path length is a measure of how well connected a network is. In the case of a weighted network higher connectivity strength will decrease the distance between two nodes, and thus a short path length indicates that, on average, any two nodes are connected by one or several strong connections. 
graphic  The cluster coefficient is defined as the number of edges connecting the neighbors of a node divided by the maximum number of edges possible between neighboring nodes. The cluster coefficient of a network is a measure of how many local clusters exist in the network. A high cluster coefficient indicates that the neighbors of a node are often also directly connected to each other, that is, they form a cluster. 
graphic  For a weighted graph, the cluster coefficient is high when the direct neighbors of a node are also interconnected and have relatively high connectivity strengths. 
graphic  The modularity score Q quantifies the degree to which a network can be divided in nonoverlapping groups. The membership of node i with a module is mi. Hence, δ(mi,mj) = 1 when two nodes i and j are in the same module and 0 otherwise. 
Measures  Interpretation 
graphic  Characteristic path length is defined as the average geodesic distance, in number of edges, connecting any two nodes in the graph, where di,j is the length of the shortest path between nodes i and j. The characteristic path length is a measure of how well connected a network is. Small characteristic path length indicates an average short distance between any two nodes, that is, they can be reached through a small number of steps. 
graphic  The weighted characteristic path length is defined as the average of the shortest paths connecting any two nodes in the graph, where wi,j is the sum of weights of the shortest weighted path between nodes i and j. The characteristic path length is a measure of how well connected a network is. In the case of a weighted network higher connectivity strength will decrease the distance between two nodes, and thus a short path length indicates that, on average, any two nodes are connected by one or several strong connections. 
graphic  The cluster coefficient is defined as the number of edges connecting the neighbors of a node divided by the maximum number of edges possible between neighboring nodes. The cluster coefficient of a network is a measure of how many local clusters exist in the network. A high cluster coefficient indicates that the neighbors of a node are often also directly connected to each other, that is, they form a cluster. 
graphic  For a weighted graph, the cluster coefficient is high when the direct neighbors of a node are also interconnected and have relatively high connectivity strengths. 
graphic  The modularity score Q quantifies the degree to which a network can be divided in nonoverlapping groups. The membership of node i with a module is mi. Hence, δ(mi,mj) = 1 when two nodes i and j are in the same module and 0 otherwise. 

Note: Parameter names (first column) are accompanied by a short explanation and an interpretation (third column). See Rubinov and Sporns (2010) for a more detailed explanation.

Based on the hypotheses that children with FLE have an abnormal brain organization caused by interference of epilepsy with normal development, and that this organization may affect cognitive function, we have analyzed cerebral functional connectivity of children with FLE in correlation to their cognitive performance using graph theoretical network parameters.

Materials and Methods

Participants

Patients with FLE were selected from our reference clinic database and were actively contacted. Inclusion criteria for the patients were a confirmed cryptogenic (i.e., presumed to be symptomatic, but with unknown etiology) localization-related epilepsy with an epileptic focus in the frontal lobe, aged between 8 and 13 years, no other disease that could cause cognitive impairment, and no history of brain injury. When patients displayed one or more episodes associated with clear ictal epileptic focal activity in the frontal lobe during EEG, the diagnosis FLE was based on EEG. When EEG was not informative, the recording of more than one seizure with clinical evidence of seizures originating from the frontal lobe was required to make the diagnosis The EEGs of all children with epilepsy were assessed by experienced clinical neurophysiologists at the Epilepsy Centre Kempenhaeghe. All FLE children had brain MRI before inclusion, which revealed no structural brain abnormalities, such as cortical dysplasia, according to the image reading of a specialized neuroradiologist.

Healthy age-matched controls were recruited by advertisements in local newspapers. Inclusion criteria were no history of brain injury or cognitive problems and visiting regular education. All subjects and parents gave written informed consent and approval for the study by the local Medical Ethical Committee was obtained.

Neuropsychological Assessment

Neuropsychological assessment was generally performed on the same day as MRI, although a small number of patients (N = 3) had their neuropsychological assessment prior (within 1 year) to MRI. Cognitive performance was measured using a computerized visual searching task (CVST) (Aldenkamp et al. 2004). This task consists of finding a grid pattern out of 24 patterns which matches the one in the center of the screen. Grid patterns are displayed in a checkerboard fashion and are numbered from 1 to 24. The target pattern is marked by an arrow on the right side and is selected by typing the correct number on the keyboard. Twenty different target patterns are presented. After 12 presentations, the surrounding grids change. The testee is asked to respond as fast as possible. The results show accuracy and speed of responses and are evaluated within the context of visual (complex) information processing and perceptual mental strategies. The most important variable indicating efficient information processing is the average reaction time.

By determining the average reaction time and the errors being made during the task, an age-corrected cognitive performance score was generated (decile score). After grouping these scores into numbers from 1 (worst score) to 10 (best score), the three worst performance scores (1, 2, and 3) were considered a manifestation of impaired cognitive performance, whereas other scores were considered normal or good.

Image Acquisition

MRI was performed on a 3.0-Tesla unit equipped with an 8-channel head coil (Philips Achieva, Philips Medical Systems). Functional MRI data were acquired using a whole-brain single-shot multislice echo-planar imaging (EPI) sequence sensitive to the blood oxygen level-dependent (BOLD) effect, with repetition time (TR) 2 s, echo time (TE) 35 ms, flip angle 90°, pixel size 2 × 2 mm2, 32 contiguous 4-mm thick slices per volume, 195 volumes per acquisition, and an acceleration factor (SENSE) of 1.5. For anatomic reference, a T1-weigthed 3D fast field echo was acquired with the following parameters: TR 8.1 ms, TE 3.7 ms, flip angle 8°, field of view 256 × 256 × 180 mm3, and voxel size 1 × 1 × 1 mm3. For safety reasons, all participants were continuously video-monitored in the magnet and were asked after scanning if they thought any seizures had happened during the procedure. We saw no evidence on video of any seizure-like responses, nor did any of the patients report having experienced a seizure. However, potential confounding influences of interical EEG abnormalities could not be ruled out as simultaneous EEG was not performed.

Network Construction

As displayed in Figure 1, data analysis sequentially consisted of the following procedures:

1) preprocessing of the measured fMRI time series; 2) anatomical parcellation and connectivity matrix; 3) network analysis; and 4) statistical analysis.

Preprocessing of Time-Series Data

The BOLD images were corrected for motion artifacts using SPM5 (Wellcome Trust Centre for Neuroimaging, UCL) software. The images were then high-pass filtered with a σ of 25 scans (50 s) and spatially smoothed (σ = 1.7 mm) using FSL 4.1.7 (Oxford University) software. Subsequently, the CSF and whole-brain signal time course were removed from the images using standard linear regression. The resulting residual time-series of the cerebrum were used for further analysis. Finally, the images were low-pass filtered (σ = 2 s, i.e., one dynamic scan interval) to remove high-frequency noise components. To assess possible confounding of motion parameters, these were compared between the groups.

Figure 1.

(A) T1-image–based subject-specific parcellation of the cortex and subcortical structures. (B) Mean time signals of the 82 parcellated regions are extracted from the regional signal fluctuations of RS-fMRI. (C) Correlation analysis is performed on each pair of time signals to construct a connectivity matrix for each subject. The colored squares indicate in which module the regions (nodes) and connections reside. (D) Modular organization of the resting-state network in the control group. In this figure, the anatomical locations of the regions are indicated as black dots. The surrounding colored circles indicate in which lobe these regions reside. The color of the lines between regions display to what modules the connections belong. It is evident that some modules occupy several lobes, while other are mainly present within one lobe (the yellow occipital module for instance).

Figure 1.

(A) T1-image–based subject-specific parcellation of the cortex and subcortical structures. (B) Mean time signals of the 82 parcellated regions are extracted from the regional signal fluctuations of RS-fMRI. (C) Correlation analysis is performed on each pair of time signals to construct a connectivity matrix for each subject. The colored squares indicate in which module the regions (nodes) and connections reside. (D) Modular organization of the resting-state network in the control group. In this figure, the anatomical locations of the regions are indicated as black dots. The surrounding colored circles indicate in which lobe these regions reside. The color of the lines between regions display to what modules the connections belong. It is evident that some modules occupy several lobes, while other are mainly present within one lobe (the yellow occipital module for instance).

Anatomical Parcellation and Connectivity Matrix

Freesurfer (Martinos Center of Biomedical Imaging) software was used to segment the T1 images of each subject into 82 cortical and subcortical regions. Freesurfer uses a surface-based alignment procedure, which might be more accurate than a volume-based alignment of a cortical atlas (Ghosh et al. 2010).

By using Matlab 7.6.0 (The MathWorks Inc.), Pearson's linear correlation coefficients between the region-averaged time-series of all pairs of Freesurfer regions were computed. In this way, for each subject, an 82 × 82 connectivity matrix was determined. This connectivity matrix included both negative and positive correlation values. The removal of the whole-brain average time-series signal tends to shift the correlation distribution to a mean value that is closer to zero, thereby creating negative correlations even if no such correlations are initially present in the data (Van Dijk et al. 2010). Only positive correlations were used for further analysis. Additionally, low (absolute) correlation coefficients could adversely affect the results as they may either represent physiologically relevant signal fluctuations or just noise. To overcome this issue, only a prespecified number of connections with highest correlation coefficients were selected, and all other connections were set to zero (Vaessen et al. 2010). Conceptually, this thresholding procedure can be expressed as a sparsity value relating the connections maintained in the network to the total number of connections possible (Achard and Bullmore 2007; Vlooswijk et al. 2011). In the remainder of this article, the results will either be presented for a particular sparsity value or as a function of sparsity.

Network Analysis

Network Characteristics

For each subject, the values of network metrics were calculated from the individual connectivity matrix. We included three widely used network metrics: characteristic path length, clustering coefficient, and modularity, by using algorithms implemented in the Brain Connectivity Toolbox (Rubinov and Sporns 2010). A description of these parameters can be found in Table 1 and in Rubinov and Sporns (2010). For these metrics, both the binary and weighted variants were calculated. For the binary metrics, the sparsity thresholded connectivity matrix was binarized by setting all edges with a correlation coefficient >0 to the value of 1. For the weighted networks, the connection matrices were divided by the mean connection weight (mean correlation coefficient over all connections), which was also evaluated separately, as this can potentially influence weighted network metrics (Ginestet et al. 2011). For the binary networks, equivalent random networks were generated (Maslov and Sneppen 2002; Vaessen et al. 2010). Clustering and path length from the random networks were compared with the measured networks to assess small-worldness (Watts and Strogatz 1998).

We compared the entire patient group with controls, the cognitively impaired patients with controls and the cognitively normal patients with the impaired patients. Between-group effects in network parameters were assessed by a two-sample (two-tailed) t-test.

Correlation Between Network Metrics and Cognitive Performance

Pearson's (linear) correlation coefficients were calculated between cognitive performance (CVST reaction time, age, and decile scores) and network parameters. We performed this analysis for the entire subject population, as well as for the patient and control groups separately.

Group Modularity

Modularity quantifies the degree to which a brain network is organized in isolated subnetworks. The more isolated the subnetworks are, the higher the modularity. We used an algorithm developed by Newman et al. (2006) to visualize the “modular structure” of the brain. With this algorithm, the brain was automatically subdivided into a number of modules (i.e., groups of nodes) with maximal correlation within and minimal correlation between the modules, creating a so-called optimal community structure (OCS) of the cerebrum. It is difficult to assess OCS at the group level, because the resulting number and spatial locations of the modules varies between subjects. For group-level analysis, a group connectivity matrix was obtained by averaging all individual correlation matrices from a subgroup of subjects (Fair et al. 2009). Another option is to concatenate the time series from all subjects, as is often done in fMRI studies using group ICA (Filippini et al. 2009), and to subsequently calculate group connectivity matrices and modularity. Both options were performed; however, interpretations of the results were similar. Therefore, only the results from the averaged connectivity matrices are reported here. The OCS for the group matrix was calculated and visualized for 1) the control group, 2) the entire patient group, and 3) the group of cognitively impaired patients.

Analysis of Individual Connections

The individual elements of the connectivity matrices were tested for group differences. Differences between the entire patient group and the control group as well as differences between the cognitively impaired patient group and the control group were assessed by mass univariate (two-sided) t-tests. Owing to the exploratory nature of this analysis, no stringent multiple comparisons methods were applied. Instead, a liberal significance threshold of P < 0.05 was used to assess possible group differences.

Furthermore, we investigated whether any aberrant connections would show a particular relation with the modularity analysis or would reveal an effect with connection length. The relation with modularity was investigated by analyzing whether differences would manifest as inter- or intramodular connections. A possible difference in anatomical orientation was investigated by measuring the angle relative to a strict left–right orientation of connections. Hence, we quantified whether aberrant edges would be oriented in an anterior–posterior of left–right orientation, disregarding their anterior–superior orientation. A possible effect of connection length was assessed by the differences in Euclidian distance between connections that showed either an increase or decrease in connection strength in the entire patient group.

Results

Preprocessing Results

The motion correction procedure was able to adequately correct for movement in the majority of the cases. The control and patient group did not differ in the amount of head movement; no significant differences were found in the mean, standard deviation, and maximum of the movement parameters. Subjects were excluded when head movements exceeded 1.5 mm/s or 1.5 °/s in at least one direction. Data of nine patients were excluded from further analysis because of movement-related artifacts (n = 6; 2 controls and 4 patients) or EPI artifacts (n = 3; 2 controls and 1 patient). The final study population for analysis included 28 patients and 37 healthy controls.

Cognition

Six patients did not complete the neuropsychological assessment and had no CVST scores. These patients are included in the group analysis, but not in the correlation analysis. In total, 11 FLE patients had a decile score below 4 and were considered cognitively impaired. The mean CVST reaction time was significantly higher in the patient group (controls: 17.3 ± 6.4 s, patients: 23.8 ± 9.5 s, P < 0.002).

Between-Group Analysis of Network Parameters

The mean matrix weight of the weighted connection matrices differ significantly neither between the control and patient group nor between the control and impaired patient group. Network parameters were assessed over a range of sparsity values (0.35–0.75). Both patient and control networks showed small-world properties indicated by a high clustering compared with equivalent random networks (mean C = 1.82, range 1.19–4.65) and a path length comparable to equivalent random networks (mean L = 1.02, range 1.00–1.09) (Watts and Strogatz 1998). The binary cluster coefficient displayed significantly higher values in the cognitively impaired patient group compared with the control group for the relatively small sparsity range 0.67–0.74. The binary path length was significantly higher in the impaired patient group compared with the control group for the narrow sparsity range 0.41–0.47. The weighted cluster coefficient was significantly higher for the impaired patient group compared with the healthy control group over the sparsity range 0.37–0.46 and 0.65–0.85. The weighted path length was significantly higher for the impaired patient group compared with the healthy control group over the sparsity range 0.64–0.74.

In contrast to L and C, we found that modularity showed significant group differences over a much wider range of sparsity values. The modularity calculated from the binary networks was higher in the entire patient group compared to the control group for the sparsity range 0.58–0.73. The impaired patient group displayed significantly higher modularity compared to the control group for the entire sparsity range. The impaired patient group also showed higher modularity scores compared to the nonimpaired patient group over the sparsity range 0.37–0.70. The weighted modularity scores were higher for both the entire patient group and the impaired patient group compared with the control group over the entire sparsity range. These results are visualized in Figure 2.

Figure 2.

The mean ± standard error (error bars) of the tested network parameters for the control group (green), the full patient group (blue), and the cognitively impaired patient group (red) as function of sparsity. An asterisk (*) indicates that the full patient group was significantly different from the control group. A hat (^) indicates that the impaired patient group was significantly different from the control group. As the sparsity increases, the number of edges in the network decreases, which causes a decrease in the binary cluster coefficient (A) and an increase in the binary path length (B) and binary and weighted modularity scores (C and F). The weighted cluster coefficient increases (D) and the weighted path length increases (E) because the remaining edges have high connection strengths and are strongly clustered (D).

Figure 2.

The mean ± standard error (error bars) of the tested network parameters for the control group (green), the full patient group (blue), and the cognitively impaired patient group (red) as function of sparsity. An asterisk (*) indicates that the full patient group was significantly different from the control group. A hat (^) indicates that the impaired patient group was significantly different from the control group. As the sparsity increases, the number of edges in the network decreases, which causes a decrease in the binary cluster coefficient (A) and an increase in the binary path length (B) and binary and weighted modularity scores (C and F). The weighted cluster coefficient increases (D) and the weighted path length increases (E) because the remaining edges have high connection strengths and are strongly clustered (D).

Network Metrics, Age, and Cognitive Performance

Within the patient group, we found that the binary modularity scores significantly increased with decreased cognitive performance (i.e., increased CVST reaction time) for all sparsity values (mean r = 0.48; range: 0.44–0.55; all P < 0.03). See Figure 3 for a plot of the correlation between CVST reaction times and modularity scores at sparsity = 0.48. The same effect was found with the CVST decile scores, which is a normalized age- and gender-corrected score (mean r = −0.55; range: −0.63 to −0.47; all P < 0.02). Higher modularity scores were associated with the longer reaction times and the lower decile scores (i.e., poor performance). The CVST reaction times positively correlated with the weighted modularity scores over the entire sparsity range (mean r = 0.47; min–max: 0.43–0.54; all P < 0.05). The CVST decile scores were negatively correlated with the weighted modularity scores over the entire sparsity range (mean r = −0.45; range −0.55 to −0.40; all P < 0.05). No significant correlations were found in the control group. No significant correlation was found between age (range 8–12 years) and any of the network parameters. The CVST reaction times and decile scores were not significantly correlated with the binary or weighted cluster coefficient or path length in the patient or control groups.

Figure 3.

Scatter plot of the modularity scores against CVST scores for the control, entire patient, and impaired patient groups. The regression line (black) of the correlation between modularity and CVST for the (entire) patient group is displayed.

Figure 3.

Scatter plot of the modularity scores against CVST scores for the control, entire patient, and impaired patient groups. The regression line (black) of the correlation between modularity and CVST for the (entire) patient group is displayed.

Visualization of Cerebral Modularity

The OCS was calculated for networks thresholded at sparsity = 0.48. About half of the possible number of edges (3500 of 6724) in the full matrix is included at this sparsity value, which reflects a good balance between the presence of noisy edges and an overly sparse matrix.

Four modules were found in all groups at this sparsity threshold. All modules were organized in a bilateral fashion. Figure 4 visualizes the organization of the OCS in the cerebrum, such that group differences in modules can be observed. Considering the control group, module 1 was located mainly in the frontal and parietal lobes (blue in Fig. 4), module 2 was mainly located in the frontal lobe (red in Fig. 4), module 3 was mainly located in the occipital lobe (yellow in Fig. 4), whereas module 4 was mainly located in the frontal and temporal lobes (green in Fig. 4). Both the patient group as a whole and the cognitively impaired patients displayed several differences in modular structure in both hemispheres in comparison to the control group. Module 4 (the green module), occupies occipital, parietal, temporal, and prefrontal regions in the control group, whereas this module curtails only to temporal regions in the impaired patient group. Consequently, module 1 (blue) occupies most of the prefrontal regions, and therefore, the diversity of the modular composition in the frontal lobe decreases in the entire patient group, and even further decreases in the cognitively impaired patient group relative to the control group (see Supplementary Fig. S1). Although the vast majority of modular differences were located in frontal regions (n= 21), as could be expected considering the frontal seizure focus in FLE, parietal (n= 8), temporal (n= 2), and occipital (n= 2) regions are also involved.

Figure 4.

In the OCS, four distinct modules of the cerebrum are visualized by different colors (as in Fig. 1) for the healthy controls (A), the full patient group (B), and the cognitively impaired patient subgroup. Module 1 (blue) extends from fronto-parietal regions in controls to more prefrontal and latero-frontal regions in patients, particularly for the cognitively impaired patients. Module 2 (red) reveals no apparent differences between patients and controls. Module 3 (yellow) extends from mere posterior occipital regions in controls to parietal and more latero-occipetal regions in patients. Module 4 (green) curtails from occipital, parietal, temporal, and frontal regions to temporal and frontal regions.

Figure 4.

In the OCS, four distinct modules of the cerebrum are visualized by different colors (as in Fig. 1) for the healthy controls (A), the full patient group (B), and the cognitively impaired patient subgroup. Module 1 (blue) extends from fronto-parietal regions in controls to more prefrontal and latero-frontal regions in patients, particularly for the cognitively impaired patients. Module 2 (red) reveals no apparent differences between patients and controls. Module 3 (yellow) extends from mere posterior occipital regions in controls to parietal and more latero-occipetal regions in patients. Module 4 (green) curtails from occipital, parietal, temporal, and frontal regions to temporal and frontal regions.

Modular- and Distance-Based Characteristics of Aberrant Connections

As shown in Figure 5A, a number of connections were found to be altered in the patient group, and a larger number of aberrant connections were found in the impaired patient group. These results were obtained at the same sparsity threshold as the one used for visualization of the modules (sparsity = 0.48). Of the connections significantly different at the P < 0.05 level, 52 (41%) were intramodular connections, whereas 151 (59%) were intermodular connections. Moreover, Figure 5B indicates that most connections weaker in patients (P < 0.05) are oriented anterior-posteriorly, when the connections stronger in the patients are mainly oriented left–right and interhemispheric. This discrepancy in orientation was quantified by measuring the angle of the connections with respect to left–right axis (a 90° angle would indicate a pure anterior–posterior orientation of an edge). This revealed that the connections weaker in patients had a significantly higher angle (hence were oriented more anterior-posteriorly) than the connections stronger in patients (65.3 ± 22.2° vs. 38.8 ± 27.5°, P< 0.001). We tested whether the anatomical distance between the connections that were either stronger or weaker would differ. We found that the connections that were weaker in the patient group were on average longer (76.8 ± 26.9 mm) than those that were stronger in the patient group (65.7 ± 26.0 mm, P < 0.003), see Figure 5C.

Figure 5.

(A) Top row: connectivity matrices for all three groups; controls (middle), patients (left), and impaired patients (right). Colors of matrix elements indicate Fisher z-transformed correlation coefficients (truncated between 0.2 and 1). The rows and columns of the matrices are sorted by the modules found in the control group as indicated by the colored squares. Bottom row shows the connections that differed at the P < 0.01 significance level (uncorrected) between the control and patient (left) and impaired patients (right). Note that most aberrant connections are intermodular. (B) Location of abnormal connections at the P < 0.05 level. Red and green lines indicate connections weaker and stronger, respectively, in patients. Red (weaker) connections are oriented in an anterior–posterior fashion, while green (stronger) connections have a left–right orientation (C). Same set of connections as in (B) versus anatomical distance. On average, the connections weaker in patients (red dots) are longer than the connections stronger in patients (green dots). Red and green diamonds and lines indicate the average (distance and difference) of the two classes of connections.

Figure 5.

(A) Top row: connectivity matrices for all three groups; controls (middle), patients (left), and impaired patients (right). Colors of matrix elements indicate Fisher z-transformed correlation coefficients (truncated between 0.2 and 1). The rows and columns of the matrices are sorted by the modules found in the control group as indicated by the colored squares. Bottom row shows the connections that differed at the P < 0.01 significance level (uncorrected) between the control and patient (left) and impaired patients (right). Note that most aberrant connections are intermodular. (B) Location of abnormal connections at the P < 0.05 level. Red and green lines indicate connections weaker and stronger, respectively, in patients. Red (weaker) connections are oriented in an anterior–posterior fashion, while green (stronger) connections have a left–right orientation (C). Same set of connections as in (B) versus anatomical distance. On average, the connections weaker in patients (red dots) are longer than the connections stronger in patients (green dots). Red and green diamonds and lines indicate the average (distance and difference) of the two classes of connections.

Discussion

Here, we have shown for the first time that a neuronal correlate for cognitive impairment exists in children with FLE. Our results suggest that functional networks in FLE are configured to have reduced connectivity between functional modules with a decline in long-range connectivity and an increase in interhemispheric connectivity. This was expressed by an increased modularity score in pediatric patients with epilepsy that was correlated with the cognitive impairment. Notably, in the cognitively impaired patients, the frontal lobe missed the characteristic module that functionally interacted with the temporal, parietal, and occipital regions as seen in the healthy controls.

Interestingly, the discrepancy in network organization found between the children with FLE (with cognitive impairments) and normal controls has previously been reported in healthy development, where a decrease in modularity and an increase in long-range connectivity was associated with normal brain maturation (Fair et al. 2009; Hagmann et al. 2010).

Our results relate to a larger body of literature on the relation between cognitive performance and large-scale connectivity, where it is suggested that higher cognitive functions are the result of interactions between systems involving numerous brain regions, instead of a direct relation between cognitive functioning and single brain regions (Bressler and Menon 2010; Menon 2011). This paradigm can also be extended to differences between healthy subjects and FLE patients. The brain and especially the frontal lobe is a highly connected structure and thus regional abnormalities might extend beyond the seizure focus and affect distant regions and connectivity to such regions. Functional MRI measurements are ultimately dependent on the synaptic and axonal configuration of the underlying neuronal ensembles. However, it is thought that, by ∼9 month of age, axonal connectivity is near complete (Conel 1947) but other mechanisms such as synaptic pruning (Huttenlocher 1979) and axonal myelination (Fields 2005) continue through young adulthood. How then do epileptic seizures (Dodrill 2002), daily AED use (Vermeulen and Aldenkamp 1995; Kuhnert et al. 2010) interact with the mechanisms of normal development and how do they eventually affect development of large-scale brain connectivity as measured with fMRI? It is likely that disturbances early in life in any of these mechanisms may have profound influences on large parts of the brain, as indicated by the whole-brain network results presented here.

Previous Findings

Several studies found functional connectivity abnormalities in epilepsy by means of calculating correlation coefficients between pairs of brain regions (Waites et al. 2006; Bettus et al. 2009; Zhang, Lu, Zhong, Tan, Liao, et al. 2009; Zhang, Lu, Zhong, Tan, Yang, et al. 2009; Pereira et al. 2010; Wang et al. 2011), but in only one study an analysis of these correlations was performed in terms of network parameters (Liao et al. 2010). Moreover, most of these studies focused on local connectivity abnormalities (only a few regions of the brain were analyzed or considered as a reference), while here, we primarily analyzed “global” brain connectivity. Cognitive functioning depends on several cerebral networks instead of isolated brain regions. It is reasonable to assume that, in patients with epilepsy, a disruption of whole-brain networks is involved in the development of cognitive deficits, instead of a localized disruption at the site of seizure focus only (Vlooswijk et al. 2011).

Increased Modularity But Preserved Small-Worldness in FLE

The means of both variants of the cluster coefficient were higher in the patients and further increased in the impaired patient group. For both variants of the path length, the same effect could be observed. The networks with high path length and high clustering are also known as regular networks (Sanz-Arigita et al. 2010): these are networks with high local clustering but few connections linking distant nodes. These findings are in line with the modularity analysis: high path length and high clustering are signs that the patient networks are organized in tightly clustered modules with only limited intermodular connectivity. However, the high clustering and comparable low path length, compared with equivalent random networks, indicate that the resting-state functional networks of both groups are still organized as an efficient small-world network (Stam et al. 2007). Previous studies have found altered small-world networks in epilepsy patients (Liao et al. 2010; Vlooswijk et al. 2011). It remains to be elucidated why these parameters only showed limited effects in this study.

The network parameter modularity did show ample significant group differences. Patients, especially the cognitively impaired patients, showed higher modularity scores than controls, suggesting the presence of more functionally isolated brain modules. In line with these findings, longer reaction times (greater cognitive impairment) correlated with higher modularity scores within the patient group. It is possible that increases in coherent activity between functional systems (integration) might facilitate particular cognitive abilities. Therefore, a reduced amount of integration could lead to an impairment of cognitive functions.

When we visualized the modular structures of different subject groups (Fig. 4), we observed a rearrangement of modular structures between controls and patients, which was more pronounced in the cognitively impaired subgroup. These findings suggest that disruptions of functional brain network modularity in children with cryptogenic FLE are related to their cognitive impairment. Furthermore, the vast majority of modular differences were located in the frontal lobe, as could be expected considering the frontal seizure focus in FLE. Importantly, module 4 (green) comprised prefrontal, temporal, parietal, and occipital regions and might thus facilitate information integration over spatially distributed regions of the brain. Especially, this module curtailed to mere temporal regions in the impaired patient group. The functional significance of this module remains to be elucidated, but the corresponding connections to the frontal lobe might be an interesting target for future studies. The finding that modular abnormalities did not seem to be restricted exclusively to the frontal lobe might imply that regions of other parts of the brain are also involved in the process that hinder some individuals with FLE to successfully perform complex cognitive tasks such as the CVST. This could also be an explanation for the broad variety of cognitive impairment seen in children with FLE.

Aberrant Connections and Anatomical Distance

We found both decreases and increases in connection strengths in the patient group (Fig. 5B), although given the large number of connections tested, these findings should be interpreted with care. A further analysis on the anatomical length and orientation of these connections did reveal an interesting effect. Connections with decreased strength in the patients were on average longer than the connections that were increased in the patients, compared with the healthy controls. Furthermore, a predominantly anterior–posterior orientation of the connections with decreased strength and a left–right (and thus interhemispheric) orientation of connections with increased strength in the patients was found. This is in agreement with prior studies that showed that in early (normal) development mainly the long-range connections increase in strength, whereas the strength of the short-range connections decreased (Fair et al. 2009; Supekar et al. 2009; Hagmann et al. 2010). This raises the question of whether FLE interferes with the normal development of functional brain networks (Power et al. 2010; Uddin et al. 2010). However, due to the narrow age range of our study population, network metrics could not be related to age. Future studies should include FLE patients with various, but well-defined, types of cognitive impairments, and age-matched healthy controls with a much wider age range to investigate whether cognitive impairment in FLE can be modeled as a developmental delay (Church et al. 2009). Furthermore, the nature of network abnormalities in different childhood neurological diseases such as epilepsy, ADHD, and autism, where cognitive impairments play a major role, should be compared. This might shed light on the commonalities and differences of network abnormalities in relation to developmental trajectories and cognitive profiles. Whether the observed changes in network organization are driven by altered developmental trajectories or by functional abnormalities in epileptic networks remain elusive. Studies using the combined EEG/fMRI analysis with well-defined epileptic zones might be able to differentiate the effects of epileptic networks on whole-brain connectivity from brain-wide abnormalities related to abnormal development.

Most connections that showed significant differences between the healthy control and patients were intermodular connections. These are connections that contribute to the overall integration of functional systems in the brain. These findings indicate that the higher modularity scores found in the patient and impaired patient group can be mainly attributed to connectional differences in those regions that connect different modules.

Clinical Significance and Future Research

Follow-up research is needed to investigate the relationship between cognition and measures of network topology, particularly for determining the prognostic value of these measures that predict cognitive progress or delay in time. Currently, no clinical tools are available that can reliably predict the long-term cognitive outcome and drug response in children with FLE. Individual connectivity maps and network analysis might eventually serve as an additional tool for the neurologist to tailor the therapeutic decision-making to the network characteristics of the impaired brain, and to balance this to seizure treatment.

Methodological Considerations

Several considerations in the choice of methodology for the current study should be discussed. First, the use of resting-state fMRI has advantages over task fMRI because the experiment is not dependent on subject compliance and ability to perform a task. This can especially be relevant in pediatric studies, where differences in developmental status and task performance are present. Here, all included subjects were video-monitored during scanning and were able to lie still with their eyes closed.

Second, the current study correlated functional connectivity to performance scores of a single cognitive task. The CVST is a computerized version of Goldstein's visual search task (Goldstein et al. 1973) and is recognized as a valid measurement for information processing speed. Information processing is a key cognitive function and is demonstrated to be a measurement that is sensitive to brain damage in general (DeMita et al. 1981), seizure effects (Aldenkamp et al. 2004), and even effects of epileptiform discharges (Aldenkamp et al. 2005, 2010). Despite these considerations in favor of the CVST, the assessment of other cognitive domains or global cognitive functioning is needed as well to reveal more specific neurocognitive correlates of functional reorganization.

Third, there are many algorithms designed to calculate the OCS (Meunier et al. 2010). Although we applied a widely used algorithm (Newman 2006), novel approaches for comparing the OCS between several different individual networks (Meunier et al. 2009), or groups of networks (Alexander-Bloch et al. 2010), have also been proposed. Other authors have suggested that future algorithms might include the concept of persistence of information flowing within modules over time (Delvenne et al. 2010), or use the concept of hierarchy, subdividing the modules into smaller modules, which can be further subdivided into smaller modules, and so on (Meunier et al. 2010). In the latter method, large modules have been described to represent consciously demanding tasks (i.e., working memory or the cognitive task from the current study), because they demand access to a more globally integrated processing system (Zeki and Bartels 1998). Smaller modules are supposed to represent automated, anatomically localized tasks (i.e., color vision or visual motion detection). Because consciously demanding tasks usually include a combination of automated tasks, algorithms that determine the modular structure on different levels might be useful in future research relating cerebral connectivity data to cognition.

In conclusion, our results show that network modularity analysis of whole-brain resting-state fMRI connectivity provides a sensitive marker for cognitive impairment in FLE. We found that the more cognitively impaired the FLE-patient is, the more isolated brain subnetworks appeared to function. Cognitively impaired patients seem to have a less efficient inter-regional transfer of information between functional networks. We suggest that abnormally interconnected functional subnetworks of the brain might underlie the cognitive problems in children with FLE.

Supplementary Material

Supplementary material can be found at: http://www.cercor.oxfordjournals.org/.

Funding

This work was supported by the Dutch National Epilepsy Foundation (NEF).

Notes

The authors thank M. Geerlings and J. Slenter for their technical assistance and E. Peeters and R. Berting for their assistance with the MRI scanning and P. van Mierlo for her assistance with neuropsychological testing and subject inclusion. Conflict of Interest: None declared.

References

Achard
S
Bullmore
E
Efficiency and cost of economical brain functional networks
PLoS Comput Biol
 , 
2007
, vol. 
3
 pg. 
e17
 
Aldenkamp
AP
Arends
J
de la Parra
NM
Migchelbrink
EJ
The cognitive impact of epileptiform EEG discharges and short epileptic seizures: relationship to characteristics of the cognitive tasks
Epilepsy Behav
 , 
2010
, vol. 
17
 (pg. 
205
-
209
)
Aldenkamp
AP
Arends
J
Verspeek
S
Berting
M
The cognitive impact of epileptiform EEG-discharges; relationship with type of cognitive task
Child Neuropsychol
 , 
2004
, vol. 
10
 (pg. 
297
-
305
)
Aldenkamp
AP
Beitler
J
Arends
J
van der Linden
I
Diepman
L
Acute effects of subclinical epileptiform EEG discharges on cognitive activation
Funct Neurol
 , 
2005
, vol. 
20
 (pg. 
23
-
28
)
Alexander-Bloch
AF
Gogtay
N
Meunier
D
Birn
R
Clasen
L
Lalonde
F
Lenroot
R
Giedd
J
Bullmore
ET
Disrupted modularity and local connectivity of brain functional networks in childhood-onset schizophrenia
Front Syst Neurosci
 , 
2010
, vol. 
4
 pg. 
147
 
Bettus
G
Guedj
E
Joyeux
F
Confort-Gouny
S
Soulier
E
Laguitton
V
Cozzone
PJ
Chauvel
P
Ranjeva
JP
Bartolomei
F
Guye
M
Decreased basal fMRI functional connectivity in epileptogenic networks and contralateral compensatory mechanisms
Hum Brain Mapp
 , 
2009
, vol. 
30
 (pg. 
1580
-
1591
)
Braakman
HM
Vaessen
MJ
Hofman
PA
Debeij-van Hall
MH
Backes
WH
Vles
JS
Aldenkamp
AP
Cognitive and behavioral complications of frontal lobe epilepsy in children: a review of the literature
Epilepsia
 , 
2011
, vol. 
52
 (pg. 
849
-
856
)
Bressler
SL
Menon
V
Large-scale brain networks in cognition: emerging methods and principles
Trends Cogn Sci
 , 
2010
, vol. 
14
 (pg. 
277
-
290
)
Centeno
M
Thompson
PJ
Koepp
MJ
Helmstaedter
C
Duncan
JS
Memory in frontal lobe epilepsy
Epilepsy Res
 , 
2010
, vol. 
91
 (pg. 
123
-
132
)
Church
JA
Fair
DA
Dosenbach
NU
Cohen
AL
Miezin
FM
Petersen
SE
Schlaggar
BL
Control networks in paediatric Tourette syndrome show immature and anomalous patterns of functional connectivity
Brain
 , 
2009
, vol. 
132
 (pg. 
225
-
238
)
Conel
JLR
The postnatal development of the human cerebral cortex: the cortex of the three-month infant
 , 
1947
Cambridge (MA): Harvard University Press
Delvenne
JC
Yaliraki
SN
Barahona
M
Stability of graph communities across time scales
Proc Natl Acad Sci U S A
 , 
2010
, vol. 
107
 (pg. 
12755
-
12760
)
DeMita
MA
Johnson
JH
Hansen
KE
The validity of a computerized visual searching task as an indicator of brain damage
Behav Res Methods
 , 
1981
, vol. 
13
 (pg. 
592
-
594
)
Dodrill
CB
Progressive cognitive decline in adolescents and adults with epilepsy
Prog Brain Res
 , 
2002
, vol. 
135
 (pg. 
399
-
407
)
Fair
DA
Cohen
AL
Power
JD
Dosenbach
NU
Church
JA
Miezin
FM
Schlaggar
BL
Petersen
SE
Functional brain networks develop from a “local to distributed” organization
PLoS Comput Biol
 , 
2009
, vol. 
5
 pg. 
e1000381
 
Fields
RD
Myelination: an overlooked mechanism of synaptic plasticity?
Neuroscientist
 , 
2005
, vol. 
11
 (pg. 
528
-
531
)
Filippini
N
MacIntosh
BJ
Hough
MG
Goodwin
GM
Frisoni
GB
Smith
SM
Matthews
PM
Beckmann
CF
Mackay
CE
Distinct patterns of brain activity in young carriers of the APOE-epsilon4 allele
Proc Natl Acad Sci U S A
 , 
2009
, vol. 
106
 (pg. 
7209
-
7214
)
Fox
MD
Raichle
ME
Spontaneous fluctuations in brain activity observed with functional magnetic resonance imaging
Nat Rev Neurosci
 , 
2007
, vol. 
8
 (pg. 
700
-
711
)
Friston
KJ
Functional and effective connectivity in neuroimaging: a synthesis
Hum Brain Mapp
 , 
1994
, vol. 
2
 (pg. 
56
-
78
)
Ghosh
SS
Kakunoori
S
Augustinack
J
Nieto-Castanon
A
Kovelman
I
Gaab
N
Christodoulou
JA
Triantafyllou
C
Gabrieli
JD
Fischl
B
Evaluating the validity of volume-based and surface-based brain image registration for developmental cognitive neuroscience studies in children 4 to 11 years of age
Neuroimage
 , 
2010
, vol. 
53
 (pg. 
85
-
93
)
Ginestet
CE
Nichols
TE
Bullmore
ET
Simmons
A
Brain network analysis: separating cost from topology using cost-integration
PLoS One
 , 
2011
, vol. 
6
 pg. 
e21570
 
Goldstein
G
Welch
RB
Rennick
PM
Shelly
CH
The validity of a visual searching task as an indicator of brain damage
J Consult Clin Psychol
 , 
1973
, vol. 
41
 pg. 
434
 
Hagmann
P
Sporns
O
Madan
N
Cammoun
L
Pienaar
R
Wedeen
VJ
Meuli
R
Thiran
JP
Grant
PE
White matter maturation reshapes structural connectivity in the late developing human brain
Proc Natl Acad Sci U S A
 , 
2010
, vol. 
107
 (pg. 
19067
-
19072
)
Harvey
AS
Hopkins
IJ
Bowe
JM
Cook
DJ
Shield
LK
Berkovic
SF
Frontal lobe epilepsy: clinical seizure characteristics and localization with ictal 99mTc-HMPAO SPECT
Neurology
 , 
1993
, vol. 
43
 (pg. 
1966
-
1980
)
Huttenlocher
PR
Synaptic density in human frontal cortex—developmental changes and effects of aging
Brain Res
 , 
1979
, vol. 
163
 (pg. 
195
-
205
)
Kuhnert
MT
Elger
CE
Lehnertz
K
Long-term variability of global statistical properties of epileptic brain networks
Chaos
 , 
2010
, vol. 
20
 pg. 
043126
 
Laskowitz
DT
Sperling
MR
French
JA
O'Connor
MJ
The syndrome of frontal lobe epilepsy: characteristics and surgical management
Neurology
 , 
1995
, vol. 
45
 (pg. 
780
-
787
)
Lawson
JA
Cook
MJ
Vogrin
S
Litewka
L
Strong
D
Bleasel
AF
Bye
AM
Clinical, EEG, and quantitative MRI differences in pediatric frontal and temporal lobe epilepsy
Neurology
 , 
2002
, vol. 
58
 (pg. 
723
-
729
)
Liao
W
Zhang
Z
Pan
Z
Mantini
D
Ding
J
Duan
X
Luo
C
Lu
G
Chen
H
Altered functional connectivity and small-world in mesial temporal lobe epilepsy
PLoS One
 , 
2010
, vol. 
5
 pg. 
e8525
 
Lorenzo
NY
Parisi
JE
Cascino
GD
Jack
CR
Jr
Marsh
WR
Hirschorn
KA
Intractable frontal lobe epilepsy: pathological and MRI features
Epilepsy Res
 , 
1995
, vol. 
20
 (pg. 
171
-
178
)
Manford
M
Hart
YM
Sander
JW
Shorvon
SD
National General Practice Study of Epilepsy (NGPSE): partial seizure patterns in a general population
Neurology
 , 
1992
, vol. 
42
 (pg. 
1911
-
1917
)
Maslov
S
Sneppen
K
Specificity and stability in topology of protein networks
Science
 , 
2002
, vol. 
296
 (pg. 
910
-
913
)
Menon
V
Large-scale brain networks and psychopathology: a unifying triple network model
Trends Cogn Sci
 , 
2011
, vol. 
15
 (pg. 
483
-
506
)
Meunier
D
Lambiotte
R
Bullmore
ET
Modular and hierarchically modular organization of brain networks
Front Neurosci
 , 
2010
, vol. 
4
 pg. 
200
 
Meunier
D
Lambiotte
R
Fornito
A
Ersche
KD
Bullmore
ET
Hierarchical modularity in human brain functional networks
Front Neuroinform
 , 
2009
, vol. 
3
 pg. 
37
 
Newman
ME
Modularity and community structure in networks
Proc Natl Acad Sci U S A
 , 
2006
, vol. 
103
 (pg. 
8577
-
8582
)
Patrikelis
P
Angelakis
E
Gatzonis
S
Neurocognitive and behavioral functioning in frontal lobe epilepsy: a review
Epilepsy Behav
 , 
2009
, vol. 
14
 (pg. 
19
-
26
)
Pereira
FR
Alessio
A
Sercheli
MS
Pedro
T
Bilevicius
E
Rondina
JM
Ozelo
HF
Castellano
G
Covolan
RJ
Damasceno
BP
Cendes
F
Asymmetrical hippocampal connectivity in mesial temporal lobe epilepsy: evidence from resting state fMRI
BMC Neurosci
 , 
2010
, vol. 
11
 pg. 
66
 
Power
JD
Fair
DA
Schlaggar
BL
Petersen
SE
The development of human functional brain networks
Neuron
 , 
2010
, vol. 
67
 (pg. 
735
-
748
)
Pravata
E
Sestieri
C
Mantini
D
Briganti
C
Colicchio
G
Marra
C
Colosimo
C
Tartaro
A
Romani
GL
Caulo
M
Functional connectivity MR imaging of the language network in patients with drug-resistant epilepsy
AJNR Am J Neuroradiol
 , 
2011
, vol. 
32
 (pg. 
532
-
540
)
Prevost
J
Lortie
A
Nguyen
D
Lassonde
M
Carmant
L
Nonlesional frontal lobe epilepsy (FLE) of childhood: clinical presentation, response to treatment and comorbidity
Epilepsia
 , 
2006
, vol. 
47
 (pg. 
2198
-
2201
)
Provini
F
Plazzi
G
Tinuper
P
Vandi
S
Lugaresi
E
Montagna
P
Nocturnal frontal lobe epilepsy. A clinical and polygraphic overview of 100 consecutive cases
Brain
 , 
1999
, vol. 
122
 
Pt 6
(pg. 
1017
-
1031
)
Rubinov
M
Sporns
O
Complex network measures of brain connectivity: uses and interpretations
Neuroimage
 , 
2010
, vol. 
52
 (pg. 
1059
-
1069
)
Sanz-Arigita
EJ
Schoonheim
MM
Damoiseaux
JS
Rombouts
SA
Maris
E
Barkhof
F
Scheltens
P
Stam
CJ
Loss of ‘small-world’ networks in Alzheimer's disease: graph analysis of FMRI resting-state functional connectivity
PLoS One
 , 
2010
, vol. 
5
 pg. 
e13788
 
Stam
CJ
Jones
BF
Nolte
G
Breakspear
M
Scheltens
P
Small-world networks and functional connectivity in Alzheimer's disease
Cereb Cortex
 , 
2007
, vol. 
17
 (pg. 
92
-
99
)
Supekar
K
Musen
M
Menon
V
Development of large-scale functional brain networks in children
PLoS Biol
 , 
2009
, vol. 
7
 pg. 
e1000157
 
Uddin
LQ
Supekar
K
Menon
V
Typical and atypical development of functional human brain networks: insights from resting-state FMRI
Front Syst Neurosci
 , 
2010
, vol. 
4
 pg. 
21
 
Vaessen
MJ
Hofman
PA
Tijssen
HN
Aldenkamp
AP
Jansen
JF
Backes
WH
The effect and reproducibility of different clinical DTI gradient sets on small world brain connectivity measures
Neuroimage
 , 
2010
, vol. 
51
 (pg. 
1106
-
1116
)
van den Heuvel
MP
Hulshoff Pol
HE
Exploring the brain network: a review on resting-state fMRI functional connectivity
Eur Neuropsychopharmacol
 , 
2010
, vol. 
20
 (pg. 
519
-
534
)
van den Heuvel
MP
Stam
CJ
Kahn
RS
Hulshoff Pol
HE
Efficiency of functional brain networks and intellectual performance
J Neurosci
 , 
2009
, vol. 
29
 (pg. 
7619
-
7624
)
Van Dijk
KRA
Hedden
T
Venkataraman
A
Evans
KC
Lazar
SW
Buckner
RL
Intrinsic functional connectivity as a tool for human connectomics: theory, properties, and optimization
J Neurophysiol
 , 
2010
, vol. 
103
 (pg. 
297
-
321
)
Vermeulen
J
Aldenkamp
AP
Cognitive side-effects of chronic antiepileptic drug treatment: a review of 25 years of research
Epilepsy Res
 , 
1995
, vol. 
22
 (pg. 
65
-
95
)
Vlooswijk
MC
Jansen
JF
Majoie
HJ
Hofman
PA
de Krom
MC
Aldenkamp
AP
Backes
WH
Functional connectivity and language impairment in cryptogenic localization-related epilepsy
Neurology
 , 
2010
, vol. 
75
 (pg. 
395
-
402
)
Vlooswijk
MC
Vaessen
MJ
Jansen
JF
de Krom
MC
Majoie
HJ
Hofman
PA
Aldenkamp
AP
Backes
WH
Loss of network efficiency associated with cognitive decline in chronic epilepsy
Neurology
 , 
2011
, vol. 
77
 (pg. 
938
-
944
)
Waites
AB
Briellmann
RS
Saling
MM
Abbott
DF
Jackson
GD
Functional connectivity networks are disrupted in left temporal lobe epilepsy
Ann Neurol
 , 
2006
, vol. 
59
 (pg. 
335
-
343
)
Wang
Z
Lu
G
Zhang
Z
Zhong
Y
Jiao
Q
Tan
Q
Tian
L
Chen
G
Liao
W
Li
K
Liu
Y
Altered resting state networks in epileptic patients with generalized tonic-clonic seizures
Brain Res
 , 
2011
, vol. 
1374
 (pg. 
134
-
141
)
Watts
DJ
Strogatz
SH
Collective dynamics of ‘small-world’ networks
Nature
 , 
1998
, vol. 
393
 (pg. 
440
-
442
)
Zeki
S
Bartels
A
The autonomy of the visual systems and the modularity of conscious vision
Philos Trans R Soc Lond B Biol Sci
 , 
1998
, vol. 
353
 (pg. 
1911
-
1914
)
Zhang
Z
Lu
G
Zhong
Y
Tan
Q
Liao
W
Chen
Z
Shi
J
Liu
Y
Impaired perceptual networks in temporal lobe epilepsy revealed by resting fMRI
J Neurol
 , 
2009
, vol. 
256
 (pg. 
1705
-
1713
)
Zhang
Z
Lu
G
Zhong
Y
Tan
Q
Yang
Z
Liao
W
Chen
Z
Shi
J
Liu
Y
Impaired attention network in temporal lobe epilepsy: a resting FMRI study
Neurosci Lett
 , 
2009
, vol. 
458
 (pg. 
97
-
101
)

Author notes

H. M. H. B. and J. S. H. contributed equally to this work.